tf.train.Checkpoint 和加载权重

Question

I'm training a model for seq2Seq using tensorflow. correct me if I'm wrong. I understood that the tf.train.Checkpoint is used to save just the checkpoint files which are only useful when source code that will use the saved parameter values is available. i would like to know how i could instatiate my model later on and load the trained weights from checkpoint in order to test it.

checkpoint_dir = 'training_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
checkpoint = tf.train.Checkpoint(optimizer=optimizer,
                                 encoder=encoder,
                                 decoder=decoder)

here is the code for training:

EPOCHS = 20

for epoch in range(EPOCHS):
  start = time.time()

  enc_hidden = encoder.initialize_hidden_state()
  total_loss = 0

  for (batch, (inp, targ)) in enumerate(dataset.take(steps_per_epoch)):
    batch_loss = train_step(inp, targ, enc_hidden)
    total_loss += batch_loss
    if batch % 100 == 0:
      print('Epoch {} Batch {} loss {}'.format(epoch + 1,batch, batch_loss.numpy()))
   
      
  # saving (checkpoint) the model every 2 epochs
  if (epoch + 1) % 2 == 0:
    checkpoint.save(file_prefix = checkpoint_prefix)

regards

Answer 1

Here is a proposed answer which suggests to use checkpoint manager.

    checkpoint = tf.train.Checkpoint(optimizer=optimizer,
                                     encoder=encoder,
                                     decoder=decoder)
    manager = tf.train.CheckpointManager(checkpoint, './tf_ckpts', max_to_keep=3)

    def train_and_checkpoint(net, manager)://Net is your custom model here and manager is managing checkpoints
          checkpoint.restore(manager.latest_checkpoint)
          if manager.latest_checkpoint:
            print("Restored from {}".format(manager.latest_checkpoint))
          else:
            print("Initializing from scratch.")
          EPOCHS = 20
        
          for epoch in range(EPOCHS):
           start = time.time()
        
           enc_hidden = encoder.initialize_hidden_state()
           total_loss = 0
        
           for (batch, (inp, targ)) in enumerate(dataset.take(steps_per_epoch)):
             batch_loss = train_step(inp, targ, enc_hidden)
             total_loss += batch_loss
             if batch % 100 == 0:
               print('Epoch {} Batch {} loss {}'.format(epoch + 1,batch, batch_loss.numpy()))
           
              
          # saving (checkpoint) the model every 2 epochs
           if (epoch + 1) % 2 == 0:
                saved_path = manager.save()
                print("Saved checkpoint for epoch {}: {}".format(int(epoch), save_path))


    //Run the above function once to save the checkpoints once.
    train_and_checkpoint(net, manager)

    //Instantiate a new model and restore the weights , start training again from last checkpoint
    opt = optimizer // the optimizer passed earlier
    net = Net() // your custom model
    
    checkpoint = tf.train.Checkpoint(optimizer=optimizer,
                                         encoder=encoder,
                                         decoder=decoder)
    manager = tf.train.CheckpointManager(checkpoint, './tf_ckpts', max_to_keep=3)
    
    train_and_checkpoint(net, manager)//it will restore weights from last checkpoint and start training again

Ref - https://www.tensorflow.org/guide/checkpoint#train_and_checkpoint_the_model